Crops identification in Kaikong River Basin of Xinjiang based on time series Landsat remote sensing images

The crops information of area and planting distribution has a great influence on the production management and policy making of the agricultural sector. Obtaining this information timely and accurately is not only the main content of agricultural remote sensing, but also the important reference for adjusting planting and estimating crop yield. At present, crop classification based on time series data mainly adopts medium and low spatial resolution images with long time series, while there are a large number of mixed pixels in low and medium spatial resolution images, which limits the classification accuracy of crops. The Normalized vegetation Index (NDVI) is mainly used in the selection of features of crop classification, while the application of other features selection is relatively few. With the rapid development of remote sensing technology, medium and high spatial resolution remote sensing data is becoming more and more abundant. How to make full use of medium and high spatial resolution images for crop classification has great research significance. This paper uses the Landsat7 ETM+ and Landsat8 OLI time series datasets of 2016 to extract crops from Kaikong River agricultural area of Xinjiang based on time-weighted dynamic time warping. It mainly includes pear, wheat, pepper, cotton and so on. According to the sample points collected in the field investigation, the sample database was established, and the NDVI values and PCA1 values of all kinds of samples were extracted at different time phases, that the standard NDVI sequences and PCA1 sequences were generated. In this paper, three classification schemes were designed and compared to explore the effect of DTW method on the recognition ability of different crops by using medium and high spatial resolution time series images, and to evaluate the time weight factor and the influence of NDVI combined with the first principal component (PCA1) on the classification results of crops. The principle of DTW algorithm classification was to calculate the distance value between the pixel sequence to be divided and the standard sequence of each crop. The smaller the distance value is, the higher the similarity between the sequences is. The crop type of the pixel to be divided was determined by comparing the distance value. However, because of its flexibility, the algorithm is prone to abnormal matching, and the introduction of time weight can limit this phenomenon very well. In this paper, DTW and TWDTW algorithms were used to classify crops based on NDVI data, and the classification accuracy of the two methods were 65.69% and 82.68%, respectively. It showed that the addition of time weight factor could effectively avoid the abnormal matching phenomenon of DTW algorithm and improve the ability of the algorithm to identify different crops. With the combination of NDVI and PCA1, the classification accuracy of TWDTW had increased by 2.61percentage points, and the phenomenon of the misclassification had significantly reduced. It explained that PCA1 could further expand the difference between crops and improve the classification accuracy. The experimental results showed that the TWDTW algorithm could obtain a satisfactory classification result in the case of less high spatial resolution data. It proved that the TWDTW algorithm has great application potential in the time of more and more high- spatial resolution images, and provides a reference for fine identification of crops based on time series data.